Mistral AI is a sophisticated language model and chat assistant. You can interact with the chatbot on the Mistral website by selecting “Talk to le Chat,” or you can download and run the model files on your own hardware for a local setup.
Mistral is described as an advanced sparse mixture-of-experts network. It boasts over 46 billion parameters, but thanks to efficient sparse processing, it only uses 12.9 billion parameters per token. This allows it to match the quality of large proprietary models like GPT-3.5+ while being significantly faster. Mistral supports long contexts in five languages and excels in tasks such as code generation and instruction following after tuning.
Developers can access the Mistral and Mixtral models via Mistral’s API or by deploying them with the vLLM open-source toolkit or other LLM managers. The model files are also available on Hugging Face.